Hello fam,
I’m exploring options to implement real-time audio transcription and note summarization in an Elixir/Phoenix-based app aimed at professionals who spend a lot of time in meetings (think: consultants, advisors, etc.). The goal is to capture live conversations (Zoom/Meet/etc.), convert them to accurate transcripts, and generate clean action items or summaries — ideally without lag.
I’m curious if anyone here has worked on similar pipelines involving:
- Streaming audio to an external service (e.g., Whisper, Deepgram)
- Efficiently handling backpressure and retries in case of failures
- Summarization pipelines (running concurrently or async via Oban, Flow, or Broadway?)
Also - as context, I’ve been building a tool (FastTrackr.AI) that helps financial advisors stay productive by automatically handling their meeting notes, tasks and CRM updates. It’s Elixir on the backend, and the concurrency model has been a huge help - but now I’m pushing into more complex use cases. I’m also looking at using AI for document processing, but that is for some other day.
I would love to receive some help regarding this…also happy to share what I’ve learned so far if it helps others. Would love to hear from anyone who’s tackled this kind of thing, or even architectural thoughts!
